Proof of Theorem 4, p. 13 — E_{k+1} − E_k ≤ (δ_{k+1} − δ_k − (α − 1)δ_k/(k + 1))(f(x_{k+1}) − f(x⋆))
ProvedHessianDamping.IPAHD.EK_succ_sub_lehessian-dampinginertial-algorithmlyapunovp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be a real Hilbert space and a convex function with a minimizer . Let , , , let , be non-negative sequences, and let be generated by (IPAHD). With as in (9) and the energy
if and holds at , then
Under the second growth condition the right-hand side is non-positive, so the energy is non-increasing.
Formalization Note. The minimizer is a point with for all . and are implicit on the page and added.
Preamble
import Mathlib import Definitions.Def_GoldenRatioVI_Shared_IsProxPoint import Definitions.Def_HessianDamping_IPAHD_Setting
Formal statement
namespace HessianDamping.IPAHD
/-- Proof of Theorem 4, p. 13: the increment of the energy under (G₂dis) at `k`. -/
theorem EK_succ_sub_le
{H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
(f : H → ℝ) (hfconv : ConvexOn ℝ Set.univ f) (hfC1 : ContDiff ℝ 1 f)
(xstar : H) (hxstar : ∀ y, f xstar ≤ f y)
(α : ℝ) (hα : 1 ≤ α)
(h s : ℝ) (hh : 0 < h) (hs : s = h ^ 2)
(β b : ℕ → ℝ) (hβ : ∀ k, 0 ≤ β k) (hb : ∀ k, 0 ≤ b k)
(x : ℕ → H) (hx : IsIPAHDRun f α s β b x)
(k : ℕ) (hk : 1 ≤ k)
(hG2 : 0 < b k * h * k - β (k + 1) - k * (β (k + 1) - β k)) :
EK f α h β b x xstar (k + 1) - EK f α h β b x xstar k
≤ (deltaK h β b (k + 1) - deltaK h β b k - (α - 1) * deltaK h β b k / ((k : ℝ) + 1))
* (f (x (k + 1)) - f xstar) := by sorry
end HessianDamping.IPAHD
Source
Attouch, Chbani, Fadili, Riahi, First-order optimization algorithms via inertial systems with Hessian driven damping, arXiv:1907.10536v2, §3.1.1, proof of Theorem 4, p. 13
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.